An experimental comparison of Deep Learning strategies for AUV navigation in DVL-denied environments

Accurate and robust navigation localisation systems are critical for Autonomous Underwater Vehicles (AUVs) in order to perform missions challenging environments. However, since the Global Positioning System (GPS) is not available underwater domain, task commonly fulfilled by integrating direct linear speed readings provided a Doppler Velocity Log (DVL) over time. As consequence, DVL failures or fallacies DVL-denied environments may arise as unexpected causes severe malfunctions of whole system. Motivated these considerations outstanding performance Deep Neural Networks (DNNs) supervised regression problems, Learning (DL) -based approach has been developed estimate vehicle’s body-frame velocity, without canonically employing measurements, Dead-Reckoning (DR) strategy. In particular, this work will describe framework, starting from data gathered AUVs National Oceanography Centre (NOC) during different field campaigns, through pre-processing inference predicted velocity. Finally, comprehensive offline comparison between DL-based models presented assess validity proposed approach.